What British Columbia Suing OpenAI and Sam Altman for Failing to Notify Police Means — September 22, 2026
· CompaniesAutomation
Radar Flash Edition: British Columbia and the Peace River South School District are suing OpenAI and Sam Altman in San Francisco over the Tumbler Ridge shooting. Their safety team detected the risk and deactivated the account, but no one called the police.
Flash Edition. The Government of British Columbia has sued OpenAI and Sam Altman in a California court: it contends that a single call to the police would have prevented the Tumbler Ridge shooting. The rest of the day's news is in today's Radar.
What happened
The province's Attorney General, Niki Sharma, announced on Monday, September 21, the lawsuit filed alongside the Peace River South School District in the San Francisco federal court against OpenAI, Altman, and associated entities (CBC, Canadian Press). On February 10, 18-year-old Jesse Van Rootselaar killed six minors and two adults in Tumbler Ridge before committing suicide.
The core issue is operational, not moral. OpenAI's safety team detected conversations about gun violence in 2025, opened an internal review, and deactivated the account—but, according to the lawsuit, they did not block the user at the identity level nor notify the RCMP, and the attacker continued using ChatGPT with a second account. Altman apologized in April for "failing to alert law enforcement"; the province cites this as an admission. OpenAI speaks of "an unspeakable tragedy" and says it continues to collaborate with governments and police, but has refused to hand over the chats.
Why it matters
The plaintiff is a government, not a family, joining thirty other active lawsuits regarding the same case. Two allegations change the framework for any company deploying AI: negligence for failing to escalate a detection and defective product. The second is the tough one: the lawsuit alleges that OpenAI configured ChatGPT to maximize conversation—weakening the rejection of dangerous premises—and that safer alternatives existed. Adjusting an assistant to "retain" users stops being a product decision and becomes a liability.
For your company
If you have an assistant serving customers or employees, the question is no longer whether the model rejects a prompt, but what happens after a detection. Three things this week:
- Write the escalation path. Who receives the alert, how quickly, and what they do. Closing the session does not close the risk: deactivating without blocking is exactly what is being judged.
- Block by person, not by session. If your identifier is the session or the email, the second account brings you back to square one.
- Request logs in the contract. Log retention, export of flagged conversations, and notification deadlines. OpenAI has refused to show chats to a government.
The unattended step of the AI First ladder is not climbed by removing the human, but by putting in writing when the human must be called.
Frequently Asked Questions
Does this affect me if I only use ChatGPT in my company?
In one respect, yes: you don't know what your provider does when their system flags one of your conversations. Ask for this in writing before renewing.
Should I notify the police if my assistant detects a threat?
It depends on the jurisdiction; in many regions, there is no general rule. That's why it matters to have a written path and a designated responsible person.
Is there a verdict yet?
No. It is a lawsuit filed on September 21, 2026, seeking a jury trial and damages exceeding $75,000, with no exact figure specified. It will take years.